In structure 1gxd, interfaces 1 and 2 should be clustered but they aren't since the rmsd must be above our 2 A threshold. PISA does not cluster them either.
The problem here is that the NCS-related chains (A,B) and (C,D) vary quite a lot in several loops, plus there's also some displacement between the 2 domains of (A,B).
A possible fix can be to measure rmsd only on contacting region.
In structure 1gxd, interfaces 1 and 2 should be clustered but they aren't since the rmsd must be above our 2 A threshold. PISA does not cluster them either.
The problem here is that the NCS-related chains (A,B) and (C,D) vary quite a lot in several loops, plus there's also some displacement between the 2 domains of (A,B).
A possible fix can be to measure rmsd only on contacting region.
See also related issues #1 and #2